Security Analysis of Cache Replacement Policies

TL;DR

This paper introduces a new model to quantify and analyze security properties of shared CPU caches, focusing on information absorption and extraction, and compares different cache replacement policies for their security implications.

Contribution

It presents a novel framework for systematically assessing cache security, including algorithms to improve static analysis bounds and compare policies like FIFO, LRU, and PLRU.

Findings

The model quantifies information absorption and extraction in caches.

Comparison of FIFO, LRU, and PLRU policies reveals their security differences.

An algorithm improves static cache analysis bounds.

Abstract

Modern computer architectures share physical resources between different programs in order to increase area-, energy-, and cost-efficiency. Unfortunately, sharing often gives rise to side channels that can be exploited for extracting or transmitting sensitive information. We currently lack techniques for systematic reasoning about this interplay between security and efficiency. In particular, there is no established way for quantifying security properties of shared caches. In this paper, we propose a novel model that enables us to characterize important security properties of caches. Our model encompasses two aspects: (1) The amount of information that can be absorbed by a cache, and (2) the amount of information that can effectively be extracted from the cache by an adversary. We use our model to compute both quantities for common cache replacement policies (FIFO, LRU, and PLRU) and to compare their isolation properties. We further show how our model for information extraction leads to an algorithm that can be used to improve the bounds delivered by the CacheAudit static analyzer.